DEVELOPMENTS
IN NONCOMMUTATIVE KOROVKIN-TYPE THEOREMSM. N. N. NAMBOODIRI
Department ofMathematics,
Cochin University ofScience and Technology, Kochi-682022. Email: nambu@cusat.ac.in,nbdri@yahoo.com
1. INTRODUCTION
The classical approximation theorem dueto P. P. Korovkin [11] in 1953,
unified many existing approximation processes such
as
Bemsteinpolyno-mial approximationofcontinuous functions. Korovkin’s discovery inspired many researchers that lead to Korovkin-type theorems and Korovkin sets in various settings such as
more
general function spaces, Banach algebras,Banachlattices and operator algebras. Anothermajor advancementwas the discoveryofgeometric theory ofKorovkinsets by Y. A.
\v{S}a\v{s}kin
[21]and D.E. Wulbert in 1968 [7]. A detailed survey of most of these developments
can
befound in the article of Berens and Lorentz in 1975 [7], monograph of Altomare and Campiti [1] most recent survey byAitomare
[2] whichcontains several
new
results also.This article aims at providing arather short survey of the developments
in the so called noncommutative Korovkin-type approximation theoly and
Korovkin sets (quantization of Korovkin theorems, W. B. Arveson [6]) in
the settings of$C^{*}$ and $W^{*}$-algebras. Due to technical
reasons
onlyimpor-tant theorems are quoted that too without proofs. However
an
attempt ismade to provideillustrative examples, afew
new
results andresearchprob-lems. First
we
quote three major theorems due to Korovkin following thearticle ofBerens and Lorentz [7]. We designate these as Korovkin’s type
I, type II and type III theorems. The survey will be about ‘quantization’ of these three theorems!
Type I Korovkin’s theorem. Let $\{\Phi_{n} : n=1,2,3, \ldots\}$ be a sequence of
positive linear maps on $C[a, b]$ and for each of the functions $g_{k}(x)=x^{k}$, $x\in[a, b],$ $k=0,1,2$, let
$\lim_{narrow\infty}\Phi_{n}(g_{k})=g_{k}$ uniformly on $[a, b],$ $k=0,1,2$
.
Then
$\lim_{narrow\infty}\Phi_{n}(f)=f$uniformlyon $[a, b]$ for all $f$ in $C[a, b]$.
Definition
1.1.
A set $S$ in $C[a, b]$ is calleda
test setor
Korovkin set forM. N. N. Namboodiri
linearoperators
on
$C[a, b],$ $\lim_{narrow\infty}\Phi_{n}(s)=s$uniformlyon
$[a, b]$ forevery
$s$ in $S$implies that $\lim_{narrow\infty}\Phi_{n}(f_{=}f$uniformly of $[a, b]$ for all $f\in C[a, b]$
.
Type I theorem says that $\{1, x, x^{2}\}$ is atest set.
Type II Korovkin’s theorem. There is not test set for $C[a, b]$ consisting
only oftwofunctions. Thus the cardinality ofatest set isatleast 3.
TypeIn Korovkin’s theorem. A triple $\{f_{0}, f_{1}, f_{2}\}$ is a test set of $C[a, b]$
exactly when itis a$\check{C}eby\check{s}ev$ system
on
$a,$$b$].
This article is divided into four sections. The next three sections
are
devoted to noncommutative Type I, Type II and Type III theorems. The last section containssome
aspects of weak Korovkin type theorems andits geometricformulation.2. TYPE I THEOREMS
This section deals with type I Korovkin theorems in the settings of $C^{*}-$
algebras. It is assumed thatthe $C^{*}$-algebras considered here
are over
com-plex numbers andalways contain identityunless otherwise specified. Deflnition 2.1. Let $d$ and $\mathscr{B}$ be complex $C^{*}$-algebras with identities $1_{d}$and $1_{\mathscr{D}}$ respectively and let $T$ : $darrow \mathscr{B}$ beapositive linear contraction (a
linear map $T$ that
preserves
positivity and such that $T(1_{d})\leq 1_{\mathscr{D}})$.
For asubset $H$ of,Ofthe Korovkin closure$K_{+}(H, T)$ is defined
as
$\{a\in d|\lim_{\alpha}\Phi_{\alpha}(a)=T(a)$ forevery net $\{\Phi_{\alpha}\}_{\alpha\in I}$ofpositivelinearcontractions from to $\mathscr{B}$ such that
$\lim_{\alpha}\Phi_{\alpha}(h)=T(h)$ forall $h$ in $H$
}
Here
convergence
considered is thenorm convergence
unless otherwise stated explicitly. Itseems
that the first noncommutative Korovkintype the-orem was due to W. B. Arveson in 1970 for $*$-homomorphisms where .Ofis $C(X)$, the $C^{*}$-algebra of all complex continuous functions on a
com-pact, Hausdorff space $X$. We recall this, being the first of its kind. Recall
that for a subset $H$ of $C(X)$, the Choquet boundary $\partial_{H}^{+}(X)$ is defined
as
the set of all points $x$ in $X$ such that the evaluation functionals $\epsilon_{x}|H$ has
the unique positive linear extension $\epsilon_{X}$ to $C(X)$. Also the support $K_{T}$ of
$T$ : $C(X)arrow \mathscr{B}$isdefined
as
setof all$x$in $X$ such that $f(x)=0$ whenever$T(f)=0$
.
2.1. Theorem. Let$H$beasubsetof$C(X)$ containing 1 andlet$T:C(X)arrow$
$\mathscr{B}$ be a $*$ homomorphism such that $K_{T}\subseteq\partial_{H}^{+}(X)$
.
Then$K_{+}(H, T)=$ $C(X)$.
Theproofoftheabovetheoremusesthelatticetheoretic properties ofthe
selfadjoint part of$C(X)$
.
It is to be recalled that the selfadjoint pall of a$C^{*}$-algebra is a lattice in the natural order if and only of.Of is
out a relation between ‘noncommutative Choquet boundary’ and
‘hyper-rigid subspaces’ (Korovkin sets) for general $C^{*}$ algebras. In whatfollows a
brief sketch of the developments from 1970 to 2010 is provided. Important theoremsthat appearedin the articles published during theabove period by
various authors, areregarding the following two questions.
(1)When does the Korovkin closure has an algebraic stmcture
(2)Whenis the Korovkin closure the full $C^{*}$ algebra. Mainly fourtypes of
maps
areconsidered in this settings. Alinearmap $\Phi$ : $arrow \mathscr{B}$is called (a) Positiveif$\Phi(x^{*}x)$ is positive for all $x\in A$(b) Schwarz
map
if$\Phi(x^{*}x)\geq\phi(x)^{*}\Phi(X)$ for all $x\in$(c) Completely positiveif$\Phi^{(n)}$ :
$\otimes M_{n}(\mathbb{C})arrow \mathscr{B}\otimes M_{n}(\mathbb{C})$ is a positive
for all positive integers$n$, where $M_{n}(\mathbb{C})$ is the setof all $n\cross n$ matrices
over$\mathbb{C}$ and $\Phi^{(n)}$ is themap on
$\otimes \mathscr{B}$defined by
$\Phi^{(n)}(a_{ij})=(\Phi(a_{ij}))$, where $(a_{ij})\in\otimes M_{n}(\mathbb{C})$
(d) Completely contractiveif$\Phi^{(n)}$ is contractive foreachpositiveinteger
$n$
.
The main toolsbeing usedinthedevelopment of the commutativetheory
for positive linearmaps areKadison-SchwarztypeinequalitiesandChoquet boundary theory. Itistobementionedthateverycompletelypositivemapof
norm
$\leq 1$ isa
Schwarzmap
and theverydefinitionofSchwarzmapimpliesthe Schwarz type inequality. Forgeneral positive linearmap $\Phi$ : $arrow \mathscr{B}$
with
norm
$\leq 1$ Kadisonproved that$\Phi(x^{2})\geq\Phi(x)^{2}\forall x\in,$$x^{*}=x$.
This fundamental inequality is known as Kadison-Schwarz inequality and
has been improved by many mathematicians like M. D. Choi [9] and T.
Fumta [10]. These improvements will have some effect on the study of
Korovkinsets. Howeverthis possibility isyet to beinvestigated.
Using Kadison-Schwarz inequality, W. M. Priestley [17] provedthe
fol-lowingtheorem in 1976.
2.2. Theorem. Let bea$C^{*}$-algebraandlet $\{\Phi_{\alpha}\}_{\alpha\in I}$ beanetofpositive
linear
maps
on
such that$\Phi_{\alpha}(1_{\ovalbox{\tt\small REJECT}})\leq 1_{\ovalbox{\tt\small REJECT}}$ $\forall\alpha\in I$.
Then the set
$J$ $:= \{x\in|\lim_{\alpha}\Phi_{\alpha}(a)=a\forall a\in\{x, x^{*}\circ x, x^{2}\}\}$
isa $J^{*}$-algebra in .
Recall that a $J^{*}$-algebra in is a norm closed, $*$ closed subset of
whichis also closed under the Jordan product $0$, namely
$a\circ b=ab+ba,$ $a,$$b\in$
M. N. N.Namboodiri
2.3.
Theorem. Let bea
$C^{*}$-algebra with identity $1_{d}$ and let $\{\Phi_{\alpha}\}_{\alpha\in I}$be
a
net of Schwarzmaps
on
such that$\Phi_{\alpha}(1_{d})\leq 1_{d}$, $\forall\alpha\in I$.
Then the subset
$K= \{x\in|\lim_{\alpha}\Phi_{\alpha}(a)=a$for$a\in\{x,$$x^{*}x,$ $xx^{*}\}\}$
is a $C^{*}$-algebra in$d$.
Subsequently B. V. Limaye and M. N. N. Namboodiri [12] improved the
results ofPriestley andRobertson toobtain the following theorem in
1982:
2.4.
Theorem. Let and $\mathscr{B}$ be $C^{*}$-algebras with identities $1_{d}$ and $1_{\mathscr{D}}$respectively. Let $\{\Phi_{\alpha}\}_{\alpha\in I}$ be
a
net of positive linear maps from $d$ to $\mathscr{B}$such that
$\Phi_{\alpha}(1_{d})\leq 1_{\ovalbox{\tt\small REJECT}}$ $\forall\alpha\in I$.
Let $T$ : $arrow \mathscr{B}$be$a*$-homomorphism. Then the subset
$J$ $:= \{x\in|\lim_{\alpha}\Phi_{\alpha}(a)=T(a)$, for$a\in\{x,$$x^{*}ox\}\}$
is a $J^{*}$-algebra in.Of.
If all $\Phi_{\alpha},$ $\alpha\in I$
are
Schwarz maps and $Ta*$-homomorphism, then $J$ isa $C^{*}$-algebra in .
In the above
cases
thetest setwas
symmetricwith respectto $*$ operation.For the
case
when this symmetry is not assumed, B. V. Limaye, M. N. N. Namboodiri in 1984 [14] and A. G. Robertson in 1986 [19] proved the followingtheorems.It is known thatKorovkin type approximation theory leads to deeper
un-derstanding ofthe stmcture under consideration. The following interesting
theoremin [14] reveals this.
2.5. Theorem. Let$=\mathscr{B}$ be
a
noncommutative $C^{*}$ algebra withidentity$1_{d}$ and let
$D= \{x\in d|\lim_{\alpha}\Phi_{\alpha}(a)=a$for all $a\in\{x,$ $x^{*}x\}\}$
where $\{\Phi_{\alpha}\}_{\alpha\in I}$ is anet ofSchwarz maps on$d$ with norm $\leq 1$. Then $D$is
a subalgebra of$d$
.
Also$D$ is $*$ closed if and only if .Of $=M_{2}(\mathbb{C})$, the setofall $2\cross 2$ matrices
over
$\mathbb{C}$.The above theorem shows that $M_{2}(\mathbb{C})$ behaves like a commutative $C^{*}-$
algebra and this is the only noncommutative one! In fact, Limaye and
Namboodiri proved that,
among
allfinite dimensional noncommutative $C^{*}$algebra $M_{2}(\mathbb{C})$ is the only one for which $D$ in Theorem 2.5 is $*$ closed.
Robertsonproved that finitedimensionalityassumption
can
bedropped.We bypass several important developments ofKorovkin approximation
theory for commutative as well
as
noncommutative Banach algebras.Ex-cellent exposition
can
be found by Micheal Panneberg, [1, Appendix $A$],Ferdinand Beckhoff [1, Appendix $B$] and F. Altomare [2]. However M.
al
so
obtains several extensionsof Korovkintypetheorems byusingoperatormonotone functions and T. Ando’s inequality. In this paperhe also unified
severalearlierresults byintroducing $0^{*}$-subalgebras and the associated
gen-eralized Schwarz maps with respect to theproduct$0$. More overtheproofs
that he gives are simplerthan the earlierones.
For the sake of completion
we
quote acouple oftheorems of Uchiyama[23] forthe $C^{*}$ algebra $C(X)$.
2.6. Theorem. [23, Theorem 3.1]
Let $S\subset C(X)$ and $C^{*}(S)$ be the $C^{*}$-algebra generated by $S$. Let $f$ be
a
operator monotone function definedon
$[0, \infty]$ such that $f(O)\leq 0$ and$f(\infty)=\infty$
.
Set $g=f^{-1}$ thenwe
have$C^{*}(S)\subseteq K_{C(X)}(SU\{g(|u|^{2}) : u\in S\})$
if$f(0)=0$ or $1\in S$. Where the set on the rightside denotes theKorovkin
closure.
2.7. Theorem. [23, Theorem 2.12] Let $\{\Phi_{n}\}$ be a sequence of Schwarz
maps from to $\mathscr{B}$ where and $\mathscr{B}$ are $C^{*}$-algebras with identities, and
let $\Phi$ : $arrow \mathscr{B}$ be
$a*$-homomorphism. Let $f$ be an operator monotone
function on $[0, \infty]$ with $f(0)=0,$ $f(\infty)=\infty$
.
Set $g=f^{-1}$. Then the set$C=\{a\in|\Phi_{n}(x)arrow\Phi(x)$ for $x=a,$$g(a^{*}a)$ and $g(aa^{*})\}$
is a $C^{*}$-subalgebra.
Another important development
was
theuse
of Krein-Millman theoremforcompact
convex
sets and the associatedunique extension property. Fornoncommutative $C^{*}$-algebras, the following theorem was proved by Taka-hasiin 1979 [22].
2.8. Theorem. Let$\phi$be
an
extremestateofa$C^{*}$ algebra andlet$x\in_{+}$peaksfor$\phi$, that is the supports of$x$ and$\phi$in the enveloping
von
Neumannalgebra add up to 1. Let $\{\phi_{\alpha}\}_{\alpha\in I}$ bea net ofpositive linearfunctionals of
and
assume
that $\lim_{\alpha}\phi_{\alpha}(1-)=1$ and $\lim_{\alpha}\phi_{\alpha}(x)=0$.
Then wehave
$\lim_{\alpha}\phi_{\alpha}(a)=0$forall $a$om .Of.
Though the above theoremiselegant, for
an
arbitrary$C^{*}$-algebras thereisno
way offinding extreme states (orpure
states), whereas
forcommutative$C^{*}$ algebras extreme states are point evaluations. In the caseof $C^{*}$-algebra
$B(H),$ $H$ aHilbert space, $\phi$a vector statebetter results
are
known to exist.See for example Limaye-Namboodiri (1979) [1, Appendix B.], Altomare
1987 [1] and Dieckmann 1992 [1].
2.9. Theorem (Altomare [1]). Let $T\in B(H)$ , be a non
zero
compactoperator, $\lambda$asimpleeigenvalue of$T$such that
$\Vert T\Vert=|\lambda|,$ $x$
a
correspondingunit eigenvectorand let $\phi=\langle\cdot x,$$x\rangle$ be thevectorstate correspondingto $x$.
Put $S=I_{H}+ \frac{1}{|\lambda|^{2}}T^{*}T-\frac{1}{\lambda}T_{\lambda}^{1}-arrow T^{*}$. Then
M. N. N.Namboodiri
Here $I_{H}$ denotes the identity operator
on
$H$.
The following theorem is due to Dieckmann
2.10. Theorem (Dieckmann). Let $T\in B(H)$ be a strictly positive
com-pact operator
on
$H$ and let $d(H)=C^{*}(I_{H}, K(H))$, where $K(H)$ is thesetofall compact operators
on
$H$. Let$\phi$bethecomplex homomorphismon
$(H)$ defined by$\phi(\lambda I_{H}+K)=\lambda,$ $K\in K(H)$. Then $K_{+}(\{I_{H}, T\}, \phi)=$
$(H)$
.
Finally we go through Arveson’s contributions to noncommutative
Ko-rovkin type theorems and Korovkin sets in 2009 [6] via his
own
theory ofnoncommutativeChoquet boundary theory of operator systems. In the fun-damental
papers
during1969-70
and2008, he introduced and provedmany
concepts and the theorems related tonon
commutative Choquet boundary and Silov boundaryideals $co$lTespondingto operator systems. This isquiteanalogous to classical theory of Choquet and Silov boundaries for
func-tion systems. Analogous to the work ofSaskin, Arveson studied the
rela-tionbetween noncommutative Korovkin setsandnoncommutativeChoquet
boundaryin2009 [6]. Heprovedmany interestingtheoremsinthis settings,
though
some
of theseresultswere
already knownto exist. Westartwith thenotion ofhyperrigid setofgenerators of$C^{*}$ algebras [6].
2.11. Definition. A finite ofcountablyinfiniteset$G$of generatorsof$C^{*}$
al-gebra.Of is said to behyperrigidiffor
every
faithfulrepresentation $\pi()\subseteq$$B(H)$ of al
on
a
Hilbertspace
$H$ and $evel\gamma$sequence
of unit preservingcompletelypositivemaps (UCP) $\Phi_{n}$ : $B(H)arrow B(H),$ $n=1,2,3,$
$\ldots$
$\lim_{narrow\infty}\Vert\Phi_{n}(\pi(g)-\pi(g))\Vert=0\forall g\in G\Rightarrow\lim_{narrow\infty}\Vert\Vert\Phi_{n}(\pi(a)-\pi(a))\Vert=0$,
$\forall a\in$
.
He then proves the followingbasic theorem.2.12. Theorem. For $evel\gamma$ separable operator system $S$ that generates
a
$C^{*}$-algebra , the following
are
equivalent.(i) $S$is hypemigid
(ii)For$evel\gamma$
non
degeneraterepresentation $\pi$ : $arrow B(H)$on
aseparableHilbert space$H$and every
sequence
$\Phi_{n}$ : $arrow B(H)$of UCPmaps;
$\lim_{narrow\infty}\Vert\Phi_{n}(s)-\pi(s)\Vert=0$ $\forall s\in S\Rightarrow\lim_{narrow\infty}\Vert\Phi_{n}(a)-\pi(a)\Vert=0$
for all $a\in d$
.
(iii)Foreverynondegeneraterepresentation$\pi$ : $arrow B(H)$ on aseparable
Hilbert space, $\pi/S$ has the unique extension property. That is, $\pi/S$ has a
uniquecompletely positive linear extensionto ,Of.
(iv) For every unital $C^{*}$ algebra $\mathscr{B}$, every unital homomorphism of $C^{*}-$
algebras $\theta$ : $arrow \mathscr{B}$andeveryUCP map $\Phi$‘ : $\mathscr{B}arrow \mathscr{B}$
$\Phi(x)=x$ $\forall x\in\theta(S)\Rightarrow\Phi(x)=x\forall x\in\theta(d)$.
One ofthemain results (Theorem 3.3 [6]) that Arveson obtainsas a
can
befound in [6]. However heproves
a $vel\gamma$ strongtheorem [6, Theorem5.1] whichis
as
follows.2.13. Theorem. Let $S$ be a separable operator system whose generated
$C^{*}$-algebra,Ofhas countablespectrum such that
$evei\gamma$irreducible
represen-tation of is aboundary representationfor$S$. Then $S$ is hyperrigid.
2.14. Arveson’s conjecture. Ifevery irreducible representation of is a boundaryrepresentation foraseparable operator system $S$,then$S$ is
hyper-rigid.
Now recall that the following theorem was proved by Y. A. Saskin for
positive linear contractions and D. E. Wulbert for linear contractions [7].
2.15. Theorem. Let $G$ be a subset of $C(X)$ that separates points of $X$
andcontains the constantfunction $1_{X}$. Then $G$is a Korovkin set for linear
contractions orpositivelinearcontractions if and only if$\partial_{Ch}G_{0}=X,$$G_{0}=$
span$G$.
The noncommutative Choquet boundary
was
defined by Arveson [6] inthefollowingway.
2.16. Definition. Let $S$ be
an
operator system in a $C^{*}$-algebra , i.e., aself adjoint linear subspace of such that $1_{d}\in S$ and $=C^{*}(S)$
the $C^{*}$-algebra generated by $S$ and $1_{d}$. A boundary representation for $S$
is
an
$i_{lT}educible$ representation $\pi$ of such that $\pi/S$ has a uniquecom-pletely positive linear extensionto $d$. The set$\partial_{S}$ of all unitary equivalence
classes of all boundary representations for $S$ id defined
as
thenoncommu-tative Choquet boundary ofthe operator system $S$.
Since$i_{lT}educible$representation of thefunction space $C(X)$ canbe
iden-tified with points in $X$ itself, Arveson’s notion of Choquet boundary for
operatorsystems is
an
exact noncommutativeanalogue of the classical oneforfunction systems.
In what follows we examinethe possibility of extending Arveson’s
theo-rem
quoted here for linearcontractions. Since extension theorem forcom-pletely positive
maps
is not available,we
needto define hypernigiditysep-arately. So
we
introduce strong hyperrigidityso as
to suitcompletely con-tractivemaps.
2.17. Definition. A finite
or
countably infinite set $G$ofgenerators ofa$C^{*}-$algebra is said to be strongly hyperrigid if for every faithful
represen-tation $\pi$ of in $B(H)$ and for every sequence $\Phi_{n}$ completely contractive
maps
from $\pi()$ to $B(H)$$\lim_{narrow\infty}\Vert\Phi_{n}(\pi(g))-\pi(g)\Vert=0$ $\forall g\in G$
M. N.N.$Nambood\ddot{m}$
2.18. Remarks. It
can
beseen
that the strong hyperrigidity coincide withhyperrigidity for UCP
as a consequence
of Arveson’sextensiontheorem for CPmaps.
Howeverhyperrigidity of$G$neednot imply strong hyperrigidity.To
overcome
this difficultyitwould be reasonabletoassume
that$G$is closedunder $*$ operation ifnecessary.
In what follows
we
aimatidentifying ‘obstmctions’ to strong hyperrigid-ity. We alsoassume
that theoperator system is $*$ closed and containsiden-tity element.
2.19. Characterisation theorem. For separable operator system $S$ that
generates a $C^{*}$-algebra ,Of, thefollowingare equivalent:
(i) $S$ is strongly hyperrigid.
(ii) For
every
non
degenerate representation $\pi$ : $arrow B(H)$on
a
sep-arable Hilbert
space
$H$ andevery sequence
$\Phi_{n}$ : $darrow B(H)$ completelycontractive maps,
$\lim_{narrow\infty}\Vert\Phi_{n}(s)-\pi(s)\Vert=0$ $\forall s\in S$
$\Rightarrow\lim_{narrow\infty}\Vert\Phi_{n}(a)-\pi(a)\Vert=0$ $\forall a\in$.
(iii) Forevery
non
degeneraterepresentation $\pi$ : $darrow B(H)$ on asepa-rable Hilbert space $H,$ $\pi/S$ has theuniqueextension property.
(iv) For$evei\gamma$ unital $C^{*}$-algebra $\mathscr{B}$, every unital homomorphism of$C^{*}-$
algebras $\theta$ : $darrow \mathscr{B}$ andevery (forevery UCPmap $\Phi$ : $\theta(d)arrow \mathscr{B}$) map
$\Phi\thetaarrow \mathscr{B}$
$\Phi(x)=x$ $\forall x\in\theta(S)\Rightarrow\Phi(x)=x$ $\forall x\in\theta(d)$.
Proof.
Theproof is more orless thesame
as thatofArveson. However thedetails
are
provided. We show that$(i)\Rightarrow(ii)\Rightarrow(iii)\Rightarrow(iv)\Rightarrow(i)$
.
$(i)\Rightarrow(ii)$
Let $\pi$ : $arrow B(H)$ be a
non
degenerate representation ofon
aseparable Hilbert space $H$ and let $\Phi_{n}$ : $darrow B(H)$ be a sequence of
(completely contractive) linearmaps such that
$\lim_{narrow\infty}\Vert\Phi_{n}(s)-\pi(s)\Vert=0$ $\forall s\in S$.
Let $\sigma$ : $arrow B(H)$ beafaithful representation of
on
another separableHilbert space $K$.
Then $\sigma\oplus\pi$ : $arrow B(K\oplus H)$ is a faithful representation of $d$
on
$K\oplus H$. Define maps $\mu_{n}$ : $(\sigma\oplus\pi)(d)arrow B(K\oplus H)$ by $\mu_{n}(\sigma(a)\oplus\pi(a))=\sigma(a)\oplus\Phi_{n}(a)$, $a\in$.
Then $\mu_{n}$ is completelycontractive.
Also $\mu_{n}(\sigma(s)\oplus\pi(s)arrow\sigma(s)\oplus\pi(s))$, for all $s\in S$.
Now,
$\lim_{narrow\infty}\sup\Vert\Phi_{n}(a)-\pi(a)\Vert\leq\lim_{narrow\infty}sub\Vert\sigma(a)\oplus\Phi(a)-\sigma(a)\oplus\pi(a)\Vert$
$=$ sub$\Vert\mu_{n}(\sigma(a)\oplus\pi(a))-\sigma(a)\oplus\pi(a)\Vert$
Therefore
$\lim_{narrow\infty}\Vert\Phi_{n}(a)-\pi(a)\Vert=0$ $a\in$.
Now $(iii)\Rightarrow(iv)$: Let$\theta$ : $arrow \mathscr{B}$ be
an
identity preserving homomorphism
of$C^{*}$-algebras andlet $\Phi$ : $arrow \mathscr{B}$ be
a
UCP that satisfies $\Phi(\theta(s))=\theta(s)$,$s\in S$
.
We have to show that$\Phi(\theta(a))=\theta(a)$ $\forall a\in$ 2.
Let $B_{0}$ bethe separable $C^{*}$-algebrain $\mathscr{B}$ generated by
$\theta()\cup\Phi(\theta())\cup\Phi^{2}(\theta())\cup\cdots$
It is clearthat $\Phi(B_{0})\subseteq B_{0}$
.
By considering afaithful representation of$B_{0}$ on a separableHilbertspace
$H$,
we
mayassume
that $B_{0}\subseteq B(H)$.
Let$\tilde{\Phi}$: $B(H)arrow B(H)$isaUCPmap
either$\overline{\Phi}/B_{0}=\Phi$
.
Here $\overline{\Phi}(\theta(s))=\theta(s),$ $\forall s\in S.$ $S$ince $\theta$ : $arrow B(H)$ isarepresentationon $H$, we must have
$\Phi(\theta(a))=\overline{\Phi}(\theta(a))=\theta(a)$ $\forall a\in A$.
Hencethe proof.
$(iv)\Rightarrow(i)$ Let $\pi$ : $arrow B(H)$ be afaithful representation of on $B(H)$
for
some
Hilbert space $H$. Put $\mathscr{B}=B(H)$. Consider the $C^{*}$-algebras ofall bounded seqences $l^{\infty}()$ and $l^{\infty}(\mathscr{B})$ in and $\mathscr{B}$ respectively. Let
$\Phi_{n}$ : $\pi(A)arrow B(H)$ be the
sequence
of all completely contractive mapssuch that
$\lim_{narrow\infty}\Vert\Phi_{n}(\pi(s)-\pi(s))\Vert=0$ $\forall s\in S$.
To show that
$\lim_{narrow\infty}\Vert\Phi_{n}(\pi(a)-\pi(a)\Vert>0$ $\forall a\in$.
Let
$\tilde{\Phi}$ :
$l^{\infty}()arrow l^{\infty}(\mathscr{B})$be defined
as
$\tilde{\Phi}(a_{1}, a_{2}\ldots a_{n}, \ldots)=(\Phi_{1}(a_{1}), \Phi_{2}(a_{2}), \ldots)$
$(a_{1}, a_{2}, \ldots, a_{n}, \ldots)\in l^{\alpha}()$
First we show that $\tilde{\Phi}$
is completely contractive. Itis quite easy to
see
that$l^{\infty}()\otimes M_{n}(C)$
can
be identified isometricallywith$l^{\alpha}(\otimes M_{n}(C))$. Thesame
way identify $l^{\infty}(\mathscr{B}\otimes M_{n}(C))$ with $l^{\infty}(\mathscr{B})\otimes M_{n}(C)$.Thus $\tilde{\Phi}^{(n)}$
can
be regarded as amap
from $l^{\infty}(\otimes M_{n}(C))$ to $l^{\infty}(\mathscr{B}\otimes$$M_{n}(C))$, foreach positive integer$n.\tilde{\Phi}^{(n)}$ is themap inducedby $\tilde{\Phi}$
for each
M. N.N.Namboodiri
It is easy to
see
that $\tilde{\Phi}^{(n)}$is contractive since $\Phi^{(n)}$ is contractive and
$\Phi^{(n)}(\pi(I))arrow\pi(I)$
as
$narrow\infty$.Let $C_{0}()$ (respectively $C_{0}(\mathscr{B})$) denotes the ideal of all
sequences
in $\mathscr{A}$(respectively $\mathscr{B}$) that
converges
tozero
innorm.
Consider themap$\tilde{\Phi}_{0}:\frac{l^{\infty}()}{C_{0}()}arrow\frac{l^{\infty}(\mathscr{B})}{C_{0}(\mathscr{B})}$
defined by
$\tilde{\Phi}_{0}(x+C_{0}())=\tilde{\Phi}(x)+C_{0}(\mathscr{B})$, $x\in l^{\infty}()$
Then $\tilde{\Phi}_{0}$ is completely contractive. Consider the embedding $\theta$ : $arrow$
$l^{\infty}(d)$ defined by
$\theta(a)=(a, a, \ldots)+C_{0}(d)$
Therefore$\tilde{\Phi}_{0}(\theta(s))=(\Phi_{1}(s), \Phi_{2}(s), \ldots)+C_{0}()$
$=(s, s, \ldots, s, \ldots)+C_{0}(d)$
$=\theta(s)$ $\forall s\in S$
Thus
$\tilde{\Phi}_{0}:\frac{l^{\infty}(d)}{C_{0}()}arrow\frac{l^{\infty}(\mathscr{B})}{C_{0}(\mathscr{B})}$
such that
$\tilde{\Phi}_{0}(\theta(s))=\theta(s)$ $s\in S$. $\Rightarrow\tilde{\Phi}_{0}$
is
a
UCP since identity $1_{d}\in S$.
This is because, if and $\mathscr{B}$are
$C^{*}-$algebras with identities $1_{d}$ and $1_{\ovalbox{\tt\small REJECT}}$ and if $\Phi$ : $darrow \mathscr{B}$ is a contractive
linear
map
such that$\Vert\Phi(1_{d})\Vert=\Vert\Phi\Vert$,
then $\Phi$ ispositivity preserving. Then $\tilde{\Phi}_{0}(\theta(a))=\theta(a)$ for all $a\in$
.
Thatis,
$(\Phi_{1}(a), \Phi_{2}(a), \ldots)+C_{0}(d)$
$=(a, a, \ldots)+C_{0}()$ $a\in$
$\Rightarrow\Vert\Phi_{n}(a)-a\Vertarrow 0$ as $narrow\infty$
$\square$
2.20. Remarks. The abovetheoremis
a
noncommutativeanalogueofWul-berts theorem for hyperrigidity in function spaces such as $C(X)$
.
This isbecause every contractive linear map on $C(X)$ is completely contractive
[3, 4]. So most of Arveson’s theorem for hyperrigidity for $C^{*}$ algebras is
valid for strong hyperrigidityalso. We state
some
ofthese without proof. 2.21. Corollary. Let $S$beastronglyhyperrigidseparable operatorsystem,with generated$C^{*}$-algebra $d_{1}$ let$K$ be
an
ideal in and let$a\in\mapsto\dot{a}\in$$d/K$be thequotientmap. Then $\dot{S}$is astronglyhyperrigid operator system
2.22. Theorem. Let $x\in B(H)$ be
a
self adjoint operator with atleast 3points in its spectmm and let be the $C^{*}$-algebra generated by $x$ and 1.
Then
(i) $G=\{1, x, x^{2}\}$ is astrongly hypeiTigid operatorsystemfor , while (ii) $G_{0}=\{1, x\}$ is nota strongly hyperrigidgenerator for.Of.
2.23. Theorem. Let $\{u_{1}, u_{2}, \ldots u_{n}\}$ be a setofisometricesthat generatea
$C^{*}$-algebra,Of and let
$G=\{u_{1}, u_{2}, \ldots u_{n}, u_{1}^{*}, u_{1}+u_{2}^{*}u_{2}+\cdots+u_{n}^{*}u_{n}\}$ .
Then $G$is astrongly hyperrigid generatorfor .
2.24. Corollary. Theset $G=\{u_{1}, u_{2}, \ldots, u_{n}\}$, where$\sum_{k=1}^{n}u_{k}u_{k}^{*}=I$, of
generators of the Cuntz algebra$\theta_{n}$ is strongly hyperrigid.
We conclude this section by remarking that many more implication of
‘
strong hyperrigiditytheorem’ are tobe investigated. However such results
will appearelsewhere.
3. TYPE II KOROVKIN THEOREMS
Recall that this section deals with thesize ofatest set $H$in a $C^{*}$-algebra
.Of generated by $H$. In $C[a, b]$, there is no test set containing only two
elements.
Observe that (i) of2.22 is already known, whereas (ii)does not seemto
existin this generality. Does this result have a noncommutativeanalogue?
3. 1. Question. Let$x\in B(H)$. Let bethe$C^{*}$-algebra generated by $I$ad
X. Then itisknown that $\{I, x, x^{*}x+xx^{*}\}$ is hyperrigid in . If spectmm
$\sigma(x)$ has atleast3 distinct points,thenisittme that $\{I, x\}$isnot
a
hyperrigidgeneratorof ?
The following simplemodification of2.22is possible.
3.2. Proposition. Let $x\in B(H)$ be normal. Then $G=\{1, x, x^{*}x\}$ is a
hyperrigid set of generators for ,Of $=C^{*}(x)$. If $\sigma(x)$ contains three
dis-tinct points $\lambda_{1},$ $\lambda_{2}$ and $\lambda_{3}$, on some straight line, then $\{$1,$x\}$ will not be a
hyperrigidgeneratorfor .
We provide the proof for the sake of completion.
Proof.
Statement (i) is already known. Now weprove
(ii) using Arveson’s argument. Let $S=$ span$\{$1,$x\}$ and let$\sigma(x)$ denote the spectmm of$x$. For$f\in C(\sigma(x))$, let $\phi_{k}(f(x))=f(\lambda_{k}),$ $k=1,2,3$.
Then $\phi_{k}$ is amultiplicative positive linear functional of
norm
1 which isan
irreducible representation of on $\mathbb{C}$. But$\phi_{k}(\lambda 1_{\alpha}+\mu_{x})=\lambda+\mu\phi_{k}(x)$
$=\lambda+\mu\lambda_{k},$ $k=1,2,3$.
But $\lambda_{2}$ is a convex combination (assume without loss of generality) of $\lambda_{1}$
M. N. N.Namboodiri
Therefore $\phi_{2}(\lambda 1_{d}+\mu x)=t\phi_{1}(\lambda 1_{d}+\mu x)+(1-t)\phi_{3}\lambda 1_{d}+\mu x)$
.
Thus the positive linear functional $\phi=t\phi_{1}+(1-\phi)\phi_{3}$ and $\phi_{2}$
are
twodifferent completely positiveextensions of$\phi_{2}/S$. Therefore the irreducible
representation $\phi_{2}$ fails to haveunique extension property therefore
one
$x$isnot hyperrigid. $\square$
Weconcludethis sectionby statinga problem of Arveson [6].
3.3. Question. Let $I=[a, b],$ $f$ : $Iarrow R$and$A\in B(H)$ be selfadjoint. Is
$[1, A, f(A)]$ hyperrigid in $C^{*}(A)$? Arveson observesthat in case $A$has
dis-crete spectmm in $[a, b]$ and if$f$is either strictly convex or strictly concave,
the
answer
is affirmative.4. TYPE III KOROVKIN THEOREMS
Recall theclassical Korovkin theorem says that $\{f_{1}, f_{2}, f_{3}\}$ is hyperrigid
in $C[a, b]$ exactly when span $\{f_{1}, f_{2}, f_{3}\}$ is a $\check{C}eby\check{s}ev$ system. It would
be interesting to examine its
non
commutative counterpart using $\check{C}eby\check{s}$ev $\vee$systems in $C^{*}$-algebra. First we recall the notion of Ceby\v{s}ev system in
Banach spaces.
4.1. Definition. Let $M$ be a subspace of a Banach space. $N$ is called a
$\check{C}eby\check{s}ev$ systemifeach vector $N$admits aunique closestpoint in $M$
.
A. Haar in 1918 [20] obtained the following characterization of finite dimensional $\check{C}eby\check{s}ev$ subspaces of $C(X),$ $X$ compact and Hausdorff. For
$C^{*}$ algebras the study
was
camiedout by A. G. Robertson, David Yost andG. K. Pederson [16]
4.2. Proposition. [7]Let$M$beann-dimensionalsubspaceof$C(X)$. Then
$M$is a $\check{C}eby\check{s}ev$ system if and only ifno non
zero
function in $M$ hasmore
that$n-1$
zeros.
4.3. Theorem. [7] Let$X$denote
an
interval $[a, b]$or
the unit circle$T$. Theneach $\check{C}eby\check{s}ev$ system $S=\{g_{0}, g_{1}, \ldots, g_{m}\}m\geq 2$ is a Korovkin set
(hy-pernigid)
It is to be remarkedthat finite Korovkinsets have been studied for
func-tion spaces, commutative Banach algebras and for
some
special types of$C^{*}$ algebras [1]. We pose the following problem whose
answer
is not yetknown.
4.4. Question. Let $M$ be
an
$n$ dimensional subspace of $C^{*}$ algebras ,where $n\geq 3$. Isit tmethat $M$is hyperrigid if itis a$\check{C}eby\check{s}ev$ subspace?
We conclude this section by mentioning few things regarding weak
Ko-rovkin type theorems.
When approximation in the weak
sense
by completely positive linear maps on $B(H)$ is considered, Korovkin type results have been obtainedin [13]. Forexample, recall the definition ofweak Korovkin set introduced in [13]. It is asfollows:
4.5. Definition. A subset $S$ of $B(H)$ is called a weak Korovkin set if for
each net $\Phi_{\alpha}$ ofcompletelypositive
maps
satisfying $\Phi_{\alpha}(I)\leq I$, therelation$\Phi_{\alpha}(s)arrow s$ weakly, $s\in S$implies $\Phi_{\alpha}(T)arrow T$ weakly $T\in B(H)$.
One of themaintheorems proved in [13] is
as
follows.4.6. Theorem. Let $S$ be
an
irreducible set $B(H)$ such that $S$ contains theidentity operator$I$ and$C^{*}(S)$ containsa
non
zerocompactoperator. Then $S$is a weakKorovkin set in $B(H)$ if and only if id$|s$ has a unique completely
positive linear extension to $C^{*}(S)$ namelyid$|_{C^{*}(S)}$.
4.7. Remarks. The condition ‘id$|s$ has a unique completely positive
lin-ear
extension to $C^{*}(S)$’ means that the identity representation of $C^{*}(S)$ isboundalyrepresentation for$S$ in the
sense
of Arveson.Thefollowingboundary theorem ofArvesonenablestoidentifyanumber
of weak Korovkin sets.
4.8. Boundary theorem ofArveson. Let $S$ be a irreducible set in $B(H)$
such that $S$ contains the identity operator and $C^{*}(S)$ contains a
non
zerocompact operator. Then the identity representation of $C^{*}(S)$ is a bound-ary representation for $S$, if and only if the quotient map $q$ : $B(H)arrow$
$B(H)/K(H)$ is not completely isometric
on
span$(S+S^{*})$ where $K(H)$denote the setof all compact operators
on
$H$.
One oftheimplications provides thefollowing example.
4.9. Example. Let$S$beanirreducible operator whichisalmost normal but
notnormal, then the set $S=\{I, s, s^{*}s+ss^{*}\}$ is a weak Korovkin set.
Acknowledgement. The author is thankful to the organisers of the RIMS
Kyoto Symposium, Oct.27-29, 2010 for the invitation
as
wellas
for localhospitality;CUSAT andNBHM-DAE, Govt. of Indiaforfinancial support.
Also the authoris thankfulto Prof. B. V.Limaye forsuggesting
some
mod-ifications and corrections.
REFERENCES
[1] F. AltomareandM. Campiti, Korovkintypeapproximationtheory anditsapplications, de Gruyter Studies inMathmatics, Berlin,NewYork, 1994.
[2] F. Altomare, Korovkin-type theorems andapproximation bypositive linearoperators,
SurveysinApproximation Theory, Vol. 5, 2010,pp. 92-164.
[3] W.B. Arveson, Subalgebras of$C^{*}$-algebras, Acta. Math. 123 (1969), 141-224.
[4] W.B. Arveson, Subalgebrasof$C^{*}$-algebrasII,Acta. Math. 128(1972),271-308.
[5] W.B.Arveson,ThenoncommutativeChoquet boundalyI,arXIV:MATH 0701329V4
1 [math OA] 19Feb 2007.
[6] W. B. Arveson, The noncommutative Choquet $boundal\gamma$ II: Hyperrigidity, ar
XIV:08102751 V4[math OA] 28May2009.
[7] H. Berens and G. G. Lorentz, Geometric theory of Korovkin sets,J. Approx.Theory,
Acad. Press 1975.
[8] J.C. Bourin and E. Rickart,Anasymmetric Kadison‘s inequality, Lin. Algebra Appl.,
433(2010)499-510.
[9] M. D. Choi,ASchwarz inequality for positive linearmapson$C^{*}$ algebras, IllinoiseJ.
M. N. N.Namboodiri
[10] T. Fumta, Asymmetnic variation of Choi’s inequality for positive linear maps,
Re-search and Its Application Of noncommutative Structure in Operator Theory,
Ab-stracts,RIMS Kyoto University Symposium,Oct27-29,2010.
[11] P. P. Korovkin, Linear operators and approximationtheory, Hindustan Publ. Corp.
Delhi, India, 1960.
[12] B. V Limayeand M. N. N. Namboodiri, Korovkin-type approximation on $C^{*}$
alge-bras,J. Approx.$TheoI\gamma,$34(1982)No. 3,237-246.
[13] B. V.Limayeand MNNNamboodiri,Weak Korovkinapproximationbycompletely
positivemapson $B[H]$,J. Approx. Theory Acad Press 1984.
[14] B. V. Limaye and MNN Namboodiri, A generalized noncommutative Korovkin theorem and closedness ofcertainsetsofconvergence,Ill. J. Math28(1984)267-280. [15] M. N. N. Namboodiri,Geometric formulationofnoncommutative (weak)Korovkin sets,Shorttalk givenatICM Satellite ConferenceonFunctional Analysis andOperator
Theory heldatISI Bangalore, August07-10, 2010.
[16] G. K.Pedersen,$\check{C}$
ebySev subspaces of$C^{*}$ algebra, Math. Scand.45, 1979, 147-156. [17] W.M. Priestley, AnoncommutativeKorovkintheorem,J. Approx. Theory, 16(1976)
No. 3,pp 251-260.
[18] A. G. Robertson, A Korovkin theorem for Schwarzmaps on$C^{*}$ algebras, Math. Z.
156,(1977)no.2,pp 205-207.
[19] A. G. Robertson, Assymmetnic invaniants setsfor completely positivemaps on $C^{*}-$ staralgebras, Bull. Austral. Math. Soc.33 [1986],no3,471-473.
[20] I. Singer: Best approximation in normed linear spaces by elements of linear subspaces, Gmndlehren d. math. Wissenschaften. 171, Springer-Verlag,
Berlin-Heidelberg-New York, 1970.
[21] Y. A.Saskin, Korovkinsystems inspacesofcontinuousfunctions,A.M. S.
Transla-tions[2],54, 1966, 125-144.
[22] S.Takahasi, Korovkin’s theorem for$C^{*}$-algebras,J. Approx. Theory, 27, 1979.
[23] M. Uchiyama, Korovkin type theorems forSchwarz maps and operator monotone